34 ms·
I don't get why you think metrics at low resolution or about resource utilization aren't usefully visualized via heatmaps. They absolutely are! Let's say your
by sciurus 8y ago
I don't get why you think metrics at low resolution or about resource utilization aren't usefully visualized via heatmaps. They absolutely are!
Let's say your application's aggregate p99 latency goes up. Generate a heatmap showing time (x-axis), percent CPU utilization (y-axis), and number of servers at that utilization (z-axis, color saturation). Oh, turns out one of your 500 applications servers has developed much higher CPU utilization than the rest! Better go kill the intern's bitcoin miner and bring latency back to normal.
http://www.brendangregg.com/HeatMaps/utilization.html http://www.brendangregg.com/HeatMaps/utilization.html
- kjeetgill 8y agoYour example is a strange one. A line graph showing time (x-axis), percent CPU utilization (y-axis) would very readily show you a single host as an outlier well above the rest.
- sciurus 8y agoA line graph like that isn't easy to render or interpret once you have many hosts. Take the example in the blog post I linked: utilization of 300 host (5,312 CPUs). Here's the data visualized as a line graph: http://www.brendangregg.com/HeatMaps/lines-allcpus-image.png http://www.brendangregg.com/HeatMaps/lines-allcpus-image.png Here it is as a heatmap: http://www.brendangregg.com/HeatMaps/cpu-utilization-heatmap-600.png http://www.brendangregg.com/HeatMaps/cpu-utilization-heatmap... Which one conveys more information?